LLM Cost Calculator
The real question is not the price per million tokens. It is what a finished task costs once a request loops, retries, and carries its context. Enter your numbers and rates below to see cost per call, per task, and per 1,000 tasks, and compare two models. Everything is editable, because rates change and only your workload is real.
Your workload
"Calls per task" is the one people forget. A request that retries or loops three times costs three times as much on the same model. See why iterations are the budget.
- Per call
- $0
- Per finished task
- $0
- Per 1,000 tasks
- $0
- Per call
- $0
- Per finished task
- $0
- Per 1,000 tasks
- $0
Monthly (30 days): $0 vs $0
Estimates only. Rates are what you enter, not quoted prices; check current provider rates. Caching, batching, and prompt trimming can change these numbers substantially.
Why cost per task, not cost per token
Pricing pages quote dollars per million tokens, which makes it easy to compare two models and easy to mislead yourself. Inside a real product a single "task" is rarely a single call. It carries a system prompt and context on every call, it often runs several times before it converges, and an agent may take many steps to finish. The honest unit is the finished task, and on that basis a cheaper-per-token model that loops twice as often is not cheaper at all.
This is also why a small, fine-tuned model can beat a large general one on cost without losing quality on a narrow task: fewer tokens, fewer retries, lower latency. The calculator makes that trade visible. If you want the deeper version, see reinforcement fine-tuning and the token economy.
How to use it
Put in the tokens a typical call sends and receives, then set how many calls it actually takes to finish one task, including retries and verification passes. Enter the input and output rates for two models you are weighing. The per-task and per-1,000-task figures are where the real decision lives, and the monthly estimate turns it into a number your finance brain understands.